1 Introduction

The deaths and disappearances of migrants while on the move are a key concern across the world. There is a serious need to investigate the factors behind it properly. Since 2014, statistics indicate that at least 4,000 people have perished every year on migration routes, and 2022 was the worst year, with more than 8,500 deaths. [Council of Europe (2024)](Migration Data Portal, 2024) In particular, children need special attention, as this demographic is particularly vulnerable due to their age. (Global Migration Data Portal, 2024)

The present study begins by illustrating the temporal and spatial distribution of these events using polar diagrams and an animated globe. Following this, an econometric panel data analysis will be employed to provide more detailed insights, to inform policy interventions to prevent future occurrences. (Thorne, 2021)

2 Methodology

The current research reports data collected from the Missing Migrants Project (International Organization for Migration, 2024)for the period from \(2014\) to \(2024\), which resulted in \(5,462\) entries after some preprocessing. Preprocessing involved the elimination of duplicates, filtering for high-quality source information, and the selection of relevant variables, identified through literature, theoretical considerations and variables available in the dataset. Furthermore, enrich the analysis with contextual factors such as country GDP and political regime. However, data on migrant populations within specific borders at given times, crucial for a robust analysis, proved challenging to obtain consistently across all regions and periods.

We transformed our variables by using dummy variables for the Cause of Death and logarithmic transformations for non-normal distributions considering skewness and kurtosis analyses. Linear relationships were incorporated into the model due to theoretical reasoning and literature on the topic. For our analysis, we employed panel data regressions, analyzing Pooled OLS, Fixed Effects, and Random Effects models. Econometric tests (including Breusch-Pagan and Hausman tests) were conducted to determine which model was most appropriate for the data. This approach allows us to isolate key factors and provide insights to inform interventions aimed at preventing future incidents.

3 Initial Analysis

The polar charts (Figure 3.1) reveal notable temporal patterns. The number of deaths shows a general increase over the years, but a decrease in \(2018\) and \(2019\). The number of missing persons presents some oscillations, with peaks around \(2016\) and \(2023\). The number of documented incidents shows a steady increase until \(2023\). Finally, the survivor counts show an overall increasing trend with a peak around \(2016\).

Figure 3.1: Polar Charts of Migrant Incidents Over Years. Distribution of Deaths, Missing, Survivors, and Incidents by Year.

The animated map (Figure 3.2) effectively shows the geographical distribution of incidents over time. It highlights the Mediterranean Sea as a critical area for migrant deaths, with notable incidents occurring in Central America and along the borders between Mexico and USA.

Figure 3.2: Animated Global Map of Missing Migrants Incidents by Year and Month.

4 Results

The panel data analysis investigates the factors influencing the number of deaths and missing children, controlling for continent and time. Table 4.1 presents the results of key statistical tests, and Table 4.2 displays the estimation results for Robust Random Effects.

The Breusch-Pagan tests, regardless of the type of model used, suggest the presence of heteroskedasticity (\(p\)-value < \(\alpha = 5\%\)). This gives statistical evidence that the errors are not homoscedastic, hence the need for the use of robust standard errors.

Lagrange Multiplier test (Breusch & Pagan, 1980) (\(p\)-value < \(\alpha = 5\%\)) supports that the Random Effects model is more appropriate than the Pooled OLS model.

For the Robust Hausman test performed, we obtained the p-value of \(0.929\). This result suggests that the Fixed and Random Effects Models are consistent, but the only the latter is efficient.

Table 4.2 presents the results from our Robust Random Effects model, which examines factors influencing child deaths or missing (D&M) in migrant incidents. Both log(Number of Females) and log(Number of Males) show positive coefficients, as an increase in overall D&M is expected to correlate with increased child D&M. However, the coefficient for log(Number of Females) is substantially larger (\(0.275\)) than that of log(Number of Males) (\(0.069\)), indicating a more pronounced effect. Specifically, a \(1\%\) increase in the number of female D&M is associated with a \(0.275\%\) increase in the number of children D&M. This suggests that women, often primary caregivers, may offer a higher degree of protection to children, and their deaths can subsequently elevate the risk for those children.

Additionally, the Drowning dummy variable has a positive coefficient of \(0.048\). This result indicates that, ceteris paribus, drowning incidents are associated with approximately a \(4.92\%\) [\(\left(e^{0.048} - 1\right) \times 100\)] increase in the expected number of children D&M compared to other causes of death.

5 Conclusion

Our analysis reveals critical patterns in migrant deaths and disappearances. Geographically, the Mediterranean, Central America, and the US-Mexico border are high-risk areas.

The panel data analysis highlights a strong link between the number of female fatalities, which revealed to be the biggest factor for child incidents in our study, and increased child deaths and disappearances, underscoring the vulnerability of women and children. Drowning incidents also significantly increase risks.